Google Algorithm Updates: Latest News and Full History

Google updated its guidance on using generative AI content on 1st Oct. The Search Central changelog logs it as "Updated guidance on using generative AI content".

The changelog's "What" line says the guide gained "information from the Search Quality Raters guidelines". Its "Why" line: "To get our documentation in sync with our presentations we use at our developer events."

Three sentences are new, in the section "Focus on accuracy, quality, and relevance". The first: generative models "don't retrieve facts, but predict a likely sequence of words based on their training data".

The second says such outputs "may contain inaccuracies (also known as hallucinations)". The third: "It is critical to manually factcheck and review all AI-generated content for accuracy and trustworthiness before publishing."

The review covers metadata too. The page says it "also applies to" title elements, meta descriptions, structured data and image alternate text, "which can appear in Search results".

The rest of the page is unchanged, by Search Engine Journal's comparison with the 10th Dec 2025 version. It still says AI used to make many pages "without adding value" may breach the scaled content abuse policy.

This is documentation, not a ranking announcement. Search Engine Journal notes that Gary Illyes said in 2024 and 2025 that AI output has to be fact-checked.

The timing falls inside the September 2026 spam update rollout. Barry Schwartz raises the link at Search Engine Land and Roundtable as a question; nothing from Google connects the two.

For publishers using AI in any part of production, the standing advice now has Google's word "critical" on it. The review step is the whole of what changed.

The changelog's RSS feed still lacked the 1st Oct item on the morning of 2nd Oct. The HTML page had it.

Seven Google researchers published a three-page paper titled "The Synthetic Gap: Automating Forensic Investigation of 'AI Slop' with the Scaled Abuse Forensics Examiner (SAFE)". Google Research's listing dates it to 2026, so this entry carries the date of Search Engine Journal's report, 25th Sep.

The problem it describes is "AI slop": "mass-produced, low-quality synthetic media designed to overwhelm traditional integrity systems". The paper calls the lag between a new generative attack and a countermeasure "the synthetic gap".

SAFE automates what a human abuse analyst does with a suspected network. A Root Agent assigns work to three specialist agents and "synthesizes these multimodal signals to render a final verdict".

A Cluster Understanding Agent maps "the relations between channels in a cluster", such as shared infrastructure. A Behavior Understanding Agent "identifies inorganic spatiotemporal patterns", such as synchronized uploads.

A Content Understanding Agent uses a LoRA-adapted language model for known violations. A few-shot model catches content that breaks "the 'spirit' of platform guidelines" without matching a known rule.

What the paper does not say matters for this hub. The words Search, website and ranking do not appear in it.

The abuse it describes is "bot-nets" of channels uploading synthetic video. Its evaluation section is written in the future tense: "We will use the following metrics".

No results are reported. The one claim of effect is the abstract's "early deployment results indicate that SAFE significantly accelerates the identification of novel synthetic threats".

Search Engine Journal asked whether SAFE is part of the September 2026 spam update. The paper gives no basis for that, and Google has not linked the two.

It is the second Google paper this year on AI-slop networks, after S-CTS in June. Together they show where Google's anti-abuse research points: coordination and behaviour across accounts, not single pieces of content.

Google released the September 2026 spam update on 24th Sep 2026 at 16:15 UTC. That is about 12:15 pm US Eastern.

The Search Status Dashboard entry reads: "Released the September 2026 spam update, which applies globally and to all languages. The rollout may take up to two weeks to complete."

It is the fourth spam update of 2026. It is also the first confirmed ranking update since the August spam update completed on 21st Aug.

The rollout window is the surprise. The three earlier 2026 spam updates were quick.

They took 19 hours 30 minutes, 2 days 1 hour and 2 days 16 hours.

Barry Schwartz asked John Mueller whether "two weeks" was a typo. Mueller replied on Bluesky: "Yes, it's likely to take longer than some of the previous ones."

Google's LinkedIn post, as quoted by Search Engine Roundtable, calls it "a normal spam update". It promises a ranking release history entry when the rollout is complete.

Search Engine Roundtable's summary of what Google told it repeats the standing lines. The update targets sites violating some of Google's spam policies.

It does not target link spam, and Google named no change from earlier spam updates.

Google would not say what share of queries it affects. It will run periodic refreshes, and recovery can take many months.

Many people reported ranking swings on 23rd and 24th Sep, before the announcement. Search Engine Roundtable's note is that "the volatility you saw prior to the announcement was not it".

Google's spam updates documentation says sites that see a change should review the spam policies. It adds that improvement can take "a period of months" while its systems re-assess a site.

If your rankings moved from 24th Sep onwards, treat it as a spam-policy question first. Work through Google's spam policies before assuming a core-update-style problem.

Our page on this update carries the rollout timeline. Google marked it complete on 8th Oct; see the completion entry.

Google announced a new search type in Search Console's performance reports on 24th Sep 2026. The Web type now splits into "Text-based" and "Multimodal".

The multimodal bucket counts "searches with Lens, Circle to Search on Android, image uploads to Google Search, and the Chrome right-click "Search this image" feature". It appears in the Search results report and in the Generative AI features report.

The rollout is global and started the same day. Google says the data shows "if your site is receiving traffic from these queries".

The data is new, not reclassified. John Mueller checked with the team and wrote: "the data wasn't previously in the counts, the report has new data."

The multimodal data is reported apart from the rest. SEO Südwest asked Mueller on 27th Sep whether it is added to the existing figures.

He replied in German that they are separate, the way Image Search is reported separately from web search.

So the text-based figures continue the old Web series, and before-and-after comparisons stay like for like.

There is no query dimension for this traffic. The help page explains that multimodal searches "mostly use images rather than text", so the queries dimension is unavailable when the type is selected.

Mueller put it the same way: "you don't see the photo", but you can see which pages showed up and how often.

The post is signed by Harsh Kharbanda, product manager lead for Google Lens, and Moshe Samet of Search Console. PPC Land notes that Search Live and image prompts inside AI Mode are not named among the four routes.

Search Engine Roundtable documented the filter before Google's post went up. It adds that the Generative AI report still carries no click or query data.

For image-heavy sites, retailers and anyone with a product catalogue, this is the first Search Console view of camera-led search. The figure to watch is the multimodal share of a page's impressions.

Google's page experience documentation gained a line on 22nd Sep, under "Page experience resources".

It reads: "CrUX ad metrics: These experimental CrUX metrics from Chrome can help you understand your site's ad count, ad density, and ad weight metrics. They provide insight into real users' ad experiences on your site, as seen and reported by Chrome."

The Search Central documentation feed had not listed the change when we checked on 23rd Sep. Search Engine Roundtable spotted it on 22nd Sep.

Barry Schwartz asked John Mueller about it. Mueller replied on Bluesky: "We included them because they're available as metrics to help evaluate the state of your site. Not everything is a direct search ranking factor!"

He added: "There are plenty of ways to improve the experience of a website, and sometimes having metrics makes it easier to discuss / work on it, "SEO" or not."

The four metrics are Chrome's, and Chrome calls them experimental. Per Search Engine Roundtable's summary, ad count is the average number of ads in the viewport.

Ad density is the share of viewport area they occupy. Ad weight is measured twice, in network bytes and in CPU milliseconds.

The document's own FAQ is unchanged on ranking. "There is no single signal. Our core ranking systems look at a variety of signals that align with overall page experience."

It still names Core Web Vitals as the metrics "used by our ranking systems". The ad metrics are not added to that answer.

So this is a documentation change, not an update. A measurement that already existed in Chrome now appears on Google Search's list of resources.

A Googler says on the record that it is not a ranking factor.

Google Research published a post on 15th Sep about Retrieve-for-Train, with a paper accepted at ICML 2026. The paper, "Efficient, Property-Aligned Fan-Out Retrieval via RL-Compiled Diffusion", has been on arXiv since 6th Mar.

Query fan-out is the technique that "breaks a single broad prompt into several related sub-queries to cover potential user interests". Google's I/O 2025 post says AI Mode "uses our query fan-out technique".

The post names two problems with using a large language model for the job at query time. The first is paraphrastic collapse, the post's term for a model producing near-synonyms instead of distinct facets.

And latency. Planning a fan-out needs "hundreds of intermediate chain-of-thought (CoT) reasoning tokens", which "creates a latency floor" at odds with "a production search bar".

Retrieve-for-Train moves the thinking offline: reinforcement learning finds fan-outs that score well on groundedness, diversity and alignment with the query. Those become training data.

A "compact, 53.9M-parameter diffusion model" then learns to produce the whole set of sub-queries in one non-autoregressive pass. The post claims a speedup of "12 to 20" times over autoregressive approaches.

The examples are shopping and fashion: "camping gear" fanning out to tents, sleeping bags, stoves and headlamps. The post does not name AI Mode, AI Overviews or Google Search.

So this is research into a mechanism Search uses, not an announcement about Search. Whether it runs in any live Google product is not stated.

For site owners the interest is the shape of the fan-out. A model trained to avoid redundant sub-queries spreads retrieval across facets of a topic, not paraphrases of the head term.

Google changed its results pages in Europe on 8th Sep to comply with the Digital Markets Act. Reuters carried the news, and Google's own verdict on it.

A Google official told Reuters the changes are the largest reduction in quality of service in the engine's 29-year history. That is Reuters' summary of the official's words.

Nick Fox, Google's senior vice-president for knowledge and information, gave Reuters a statement. The changes "degrade the user experience for Europeans", he said, "boosting online intermediaries at the expense of local businesses".

He added: "Users outside the EU will not be impacted by these changes."

What changes is the page for hotel, flight, restaurant and product searches. Reuters describes one specialised search engine highlighted at the top, then two more with fewer details.

A carousel of hotels, airlines or restaurants sits below them. Reuters says features such as real-time prices are stripped out of it.

Google's view is that the format favours comparison sites such as Expedia or Booking.com. Reuters calls these vertical search services, or VSS.

Google told Reuters that earlier DMA changes cut free, direct booking traffic to European businesses by 30%. It said tests with millions of European users showed dissatisfaction, with people retyping queries.

The context is the Commission's 460 million euro fine of 23rd Jul for Search self-preferencing. Our AI Overviews page covers it, and the Commission gave Google 60 days to comply.

Continued non-compliance can draw periodic penalties of up to 5% of worldwide turnover.

The same day, Search Central documentation gained a page on regional differences in the Search experience. Its changelog dates the addition to 8th Sep.

The page lists each feature by region and query type. The features are aggregator units, supplier units, an ecosystem carousel, job and places site features, and structured data carousels.

Türkiye and South Africa have their own rows.

Google says the page exists "to help publishers, businesses, and aggregators learn about the different regional search features available and understand the eligibility criteria and how to participate".

This is a layout change, not a ranking update. The status dashboard shows nothing, and no ranking system is named.

It is the second Search change in a fortnight that gives EEA searchers a different page from everyone else. The site reputation split of 30th Aug was the first.

A rank tracker measuring from one region sees only its own version of the page. For hotel, travel and product queries, the EEA version now leads with comparison sites.

Google announced on 28th Aug that it is changing how it enforces the site reputation abuse policy. That policy bans publishing third-party content on a trusted site to borrow its ranking power.

The practice is better known as parasite SEO.

From 30th Aug, a manual action under the policy splits by where the searcher is. Outside the EEA it works as before.

The affected portion of the site is demoted in results. The rest of the site is untouched.

Inside the EEA, "the impact of the manual action won't apply". Instead, the affected section "may be separated in our systems so that, over time, it ranks independently from the rest of the site".

Google's documentation changelog dates the updated policy page to the same day. It explains what that separation means: the pages are ranked "on their own merits".

In Google's words, "content is ranked in a consistent way against other content of the same nature (for example, casino content ranks against other casino content)".

The reason is regulatory. Google says the change follows "discussion with the European Commission", and states its objection in the same post: "we remain concerned that an overbroad application of the DMA could prevent us from addressing real threats to the integrity of our search results".

A Google spokesperson went further in a statement to Search Engine Roundtable: "Our European users are no less frustrated by parasite SEO and other deceptive, pay-to-play tactics that degrade search results, and we stand by our Site Reputation Policy."

Search Console mechanics stay in place, with one addition. Sites are still notified of a manual action and can still file a reconsideration request.

Eligible sites can now also "bring disputes to mediation" after that request. No Google manual action has had that route before.

For the first time, a Google spam policy gives the same page different outcomes by the searcher's region. Search Engine Roundtable's coverage makes that point plainly.

Google introduced the policy in March 2024 alongside that month's core and spam updates. Search Engine Roundtable notes that enforcement in Europe began in January 2025.

The practical effect for anyone running or renting space on a large site is simple. A section under a manual action can now rank normally for EEA searchers while staying demoted everywhere else.

The same page can therefore look penalized or clean depending on where you check from. A rank tracker measuring from one region only ever sees half of that.

This page exists because Google names some of its ranking updates. It is useful to know what Google thinks that naming is for.

On 27th Aug, SEO consultant Shahram Rahbari asked John Mueller on Bluesky whether naming an update publicly is "meant to give site owners any actionable signal at the page or site level, or is it purely a timing/transparency notice with no specific action expected from webmasters?"

Mueller answered the same evening: "This is primarily transparency for site owners. If there's something actionable for site owners to do, we'll publish that in our policies, help center, documentation, and blog independently -- usually well in advance of any changes in our systems."

Two things follow for anyone reading a dashboard row. A named update is a timing notice.

It tells you when to look at your analytics, and nothing about what to change.

And the instruction, when there is one, comes through a different channel and lands first. The site reputation policy change for the EEA the day after Mueller's post is the shape.

It arrived as a blog post, a documentation change and advance notice, all separate from any dashboard entry.

That ordering is the practical part. If you wait for a named update to tell you what to do, you are late.

The guidance has already been published somewhere else, usually weeks earlier.

One claim in the coverage is not Google's. Search Engine Roundtable's write-up adds that "Google does not announce all updates, not even smaller core updates".

Mueller's post does not say that. Google's own core-update documentation says the reverse for core updates: "Several times a year, Google makes significant, broad changes to our search algorithms and systems. We refer to these as core updates, and we give notice when they happen on our list of Google Search ranking updates."

The wider point underneath it is not in dispute. Google has always said it makes many ranking changes it never names.

That is why unexplained ranking swings with no dashboard row stay unpublished on this page.

Hover a result in Google Search and the status bar may now show a google.com/goto address. It is no longer the address of the page you are about to visit.

A server-side redirect now sits between the results page and the destination. Search Engine Roundtable reported on 26th Aug that Google has confirmed the change is rolling out.

A Google spokesperson gave that site a short statement. It does not name what Google is defending against: "We have a long history of deploying technical measures against evolving forms of abuse, and we regularly take steps to protect our services and users."

Search Engine Land, carrying the same confirmation, describes the link shape. The click goes to google.com/goto?url= plus a hashed form of the destination URL.

The redirect then delivers the page.

The measurement side is where this lands for anyone doing SEO. That evidence comes from a provider rather than from Google.

Derek Perkins of the rank-tracking firm Nozzle posted on 26th Aug. Google has been testing "anti-bot measures that eliminate all client side links" in favour of the goto redirects.

The first major test was widely reported in the first week of August. By that morning he was seeing "near 100% rollout across several residential ip providers".

His chart covers roughly four months of the test. Nozzle has been discarding the affected results pages "so that our customer data isn't polluted with opaque links".

The practical problem he describes is one of cost rather than possibility. "The goto links can't be decoded, so providers will have to follow the redirect links", which works at small scale, and "each SERP will have hundreds of links to decode".

He compares it to the num=100 change of autumn 2025. That is the closest precedent for a Google change that broke third-party rank data without touching rankings.

Two things are not established, and both matter before anyone acts on this.

Google's confirmation names abuse and nothing else. The widely repeated reading that this targets AI companies scraping results is other people's inference, not Google's stated reason.

Search Engine Land draws that inference in its own voice. It writes that the change "should make it harder for third-party tools, services, scrapers, and AI engines to scrape the Google search results".

It also points at SerpAPI, a scraping firm it notes Google sued and recently lost a motion against. That is the outlet's technical reading, and Google has not said it.

We have also seen no evidence on whether links inside AI Overviews and AI Mode are affected. The same goes for whether anything changes in the referrer a site receives.

Neither the statement nor the rollout reports address either question.

Search Console is the same kind of open question. Search Engine Land writes that "you have to assume" its data is unaffected, and an assumption is all that is.

Google marked the August 2026 spam update complete on the Search Status Dashboard on 21st Aug 2026. The time was 08:49 UTC, about 4:50 am US Eastern.

Measured from its 16:27 UTC start on 18th Aug, the rollout ran 2 days, 16 hours and 22 minutes. Google's own wording on the incident is simply "The rollout was complete as of August 21, 2026."

That is the longest of the three spam updates Google has shipped in 2026. It beat the June update's 2 days 1 hour by roughly 15 hours.

The March update's 19 hours 30 minutes is still the fastest rollout the dashboard has ever recorded.

It is nowhere near the longest on record. The August 2025 spam update took 26 days 15 hours.

So the 2026 pattern still holds: spam updates now land inside a few days rather than a few weeks.

What "complete" does and does not mean is the part to get right before you read your analytics.

It means the rollout Google announced has finished deploying. Ranking movement caused by this update should have settled.

It does not mean the systems behind it have stopped. Google told Search Engine Roundtable that it will run periodic refreshes of the spam update.

Those refreshes can move rankings again without any new announcement.

Google also repeated its standing position that recovery from a spam demotion can take many months. There is no reconsideration request for an algorithmic demotion.

If you lost rankings between 18th and 21st Aug, the useful next step is narrower than a general content audit. Google ruled out two suspects when it released this one.

It does not target link spam, and it has nothing to do with the site reputation abuse policy.

That points at the rest of the spam policies. Hacked or scraped content, doorway pages, expired-domain abuse and scaled content abuse are the categories that catch otherwise legitimate sites.

Our page on this update carries the full rollout timeline.

Google released the August 2026 spam update on 18th Aug 2026 at 16:27 UTC. It was Google's first confirmed ranking update since the June spam update completed on 26th Jun.

The Search Status Dashboard entry reads: "Released the August 2026 spam update, which applies globally and to all languages. The rollout may take a few days to complete."

Google's own announcement calls it "a normal spam update". It finished rolling out on 21st Aug at 08:49 UTC, after 2 days 16 hours.

That makes it the longest of the three spam updates Google has run in 2026.

Many people reported ranking swings in the days before the announcement. John Mueller settled the question on LinkedIn: "We don't roll them out beforehand. We try to get the announcement as close as possible to the actual "button-pushing" (which is sometimes challenging due to time-zones)."

Search Engine Roundtable collected the reply and adds the qualification that matters. Google does run smaller core updates without announcing them.

So movement before 18th Aug was real, it just was not this update.

Google told Search Engine Roundtable that the update targets sites violating some of its search spam policies. It also confirmed two things the update does not do.

It does not target link spam, and it is unrelated to the site reputation abuse policy.

Google declined to say what share of queries or searches the update affects. It also said it will run periodic refreshes of the spam update.

Google repeated its standing guidance that recovery from a spam demotion can take many months.

This is the third spam update Google has announced in 2026.

If your rankings moved from 18th Aug onwards, treat it as a spam-policy question rather than a content-quality one. Work through Google's spam policies first.

Look for hacked or scraped content, doorway pages, expired-domain abuse and scaled content abuse before assuming a core-update-style problem. Our full page on this update carries the day-by-day rollout timeline and the completion date.

Google released the June 2026 spam update on 24th Jun and marked it complete on 26th Jun. The 2-day, 1-hour rollout continues the much faster spam deployments Google has shipped since March.

Google described it as "a normal spam update" applying globally and to all languages. It confirmed the update does not specifically target link spam or the site reputation abuse policy.

Sites that were demoted should expect recovery, if earned, to take months rather than days.

Four Google researchers published "Scalable Detection of Adversarial Synthetic Slop and Coordinated Media Abuse: A LoRA-Enabled Multimodal Defense System". The PDF is undated, so this entry carries the date of Search Engine Journal's report, 19th Jun.

The system is "designed for online video platforms (OVP)". It identifies and terminates "clusters of coordinated accounts exhibiting a prevalence of adversarial synthetic content".

Two components do the work. A Coordinated Bot-Net Detector links accounts by relatedness signals, and a Synthetic Pattern Classifier scores their content.

The classifier adapts a large model, with "Gemini 2.0 Flash" named as the example, through LoRA and automatic prompt optimisation. The authors say they "can retrain a LoRA adapter rapidly when a new GenAI model (like Sora or Kling) is released by attackers".

Enforcement thresholds are "set high to yield high precision (92% to 95%)". The cluster requirement "acts as a safeguard": the target is produced behaviour across accounts, not one piece of content.

For text, the paper notes that "text embeddings generated by models like Sentence-BERT are used to detect scripted AI narratives". That is the passage Search Engine Journal read as a hint about web spam.

The paper does not mention Search, websites or ranking. Its PDF gives no count of clusters or channels removed.

Figures of 50,000 clusters and 130,000 channels circulate in coverage of this paper, but they are not in the PDF.

Google Research's page for the paper was already gone when the Wayback Machine captured it on 21st Aug. So we do not print those figures.

The results claim is one sentence: "Test data demonstrates the system's significant impact, resulting in the successful termination of clusters at a high precision".

Google's September paper on SAFE builds on this one. Both are about video channels, and neither is evidence about how web spam is detected.

The May 2026 core update began on 21st May and completed on 2nd Jun after 11 days and 21 hours. Search Engine Roundtable called it "a big one, causing a lot of search ranking volatility".

Tracker spikes came around 23rd May, 30th May and the final 24 hours of the rollout.

It was a standard broad core update, the second of 2026. Google pointed to its usual guidance on core updates rather than any specific target.

The March 2026 core update began on 27th Mar, three days after the March spam update completed. It finished on 8th Apr, after 12 days and 4 hours.

During the rollout, Google's John Mueller addressed why ranking swings come in waves: "We generally don't announce 'stages' of core updates... sometimes they have to work step-by-step, rather than all at one time". He added that there is no single "core update machine" being switched on.

Google's March 2026 spam update completed in 19 hours and 30 minutes. That is the fastest rollout of any update documented on the Search Status Dashboard since it launched.

For comparison, the August 2025 spam update took over 26 days.

The speed was widely read as evidence that Google's spam systems can now deploy globally in a single push. In 2024 and 2025, spam rollouts took weeks.

Google completed the February 2026 Discover core update on 27th Feb after roughly three weeks.

It is the first core update scoped to Discover rather than Search. Per Google, it was "released... for English language users in the US (will look to expand it to all countries and languages in the future)" and "is designed to improve the quality of Discover overall."

As of early July 2026 Google has not announced an expansion beyond US English.

The December 2025 core update rolled out from 11th Dec to 29th Dec, 18 days and 2 hours. It closed a year with three core updates (March, June, December) and one spam update (August).

Google flagged at release that the rollout "may take up to 3 weeks", and it nearly did.